A Study Of Mechanical Fault Diagnosis Of Diesel Engine Based On Support Vector Machine

Statistics Learning Theory (SLT) is a machine learning method based on solid theory, which is developed from traditional statistics and turns to be sophisticated system info ---- Statistics Learning Theory since 90's in 20 century. SLT provides a new pattern recognizing method --- Support Vector Machine (SVM). SVM is a promising direction in the field of machine learning, which integrated other criterion technologies of machine learning and is of unique advantage. It shows excellent performance in situations where the sample sizes are small, the sample dimensions are high and the problems are nonlinear.With the fast development of industry and technology, the construct of the modern equipment become more and more complex. There are close relations among different equipments. It bring exigent request to engineering diagnosis. The development of electron technology, especially of the computer technology, offers essential technique foundation to intelligent diagnosis. Intelligent diagnosis become a important aspect of engineering diagnosis.The paper reviews the principles of SVM, and study the influences of the kernel parameters and error penalty parameter on SVM generalization ability. Then the paper introduce three kind of method for selecting the parameters of SVM, and discusses the advantages and shortcomings. the paper presents a general overview of existing representative methods for multi-category support machines and systematically compares their performances, including training speed, classification speed and generalization ability. The disadvantage and unsolved problem of these methods are also given. At last, the paper studies the application of multi-category support machines in the fault diagnosis of diesel engine. The result of experiment shows that the SVM method has good classification ability and high efficiency for multi-fault classification in mechanical systems.